Published November 15, 2024 | Version v1
Journal article Open

Generative adversarial networks accurately reconstruct pan-cancer histology from pathologic, genomic, and radiographic latent features

  • 1. University of Chicago
  • 2. Geisinger Cancer Institute
  • 3. SimBioSys
  • 4. University of North Carolina at Chapel Hill
  • 5. University of Pennsylvania Health System
  • 6. NorthShore University HealthSystem

Description

Artificial intelligence models have been increasingly used in the analysis of tumor histology to perform tasks ranging from routine classification to identification of molecular features. These approaches distill cancer histologic images into high-level features, which are used in predictions, but understanding the biologic meaning of such features remains challenging. We present and validate a custom generative adversarial network—HistoXGAN—capable of reconstructing representative histology using feature vectors produced by common feature extractors. We evaluate HistoXGAN across 29 cancer subtypes and demonstrate that reconstructed images retain information regarding tumor grade, histologic subtype, and gene expression patterns. We leverage HistoXGAN to illustrate the underlying histologic features for deep learning models for actionable mutations, identify model reliance on histologic batch effect in predictions, and demonstrate accurate reconstruction of tumor histology from radiographic imaging for a "virtual biopsy."

Data availability

All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Data from TCGA including digital histology and most of the clinical annotations used are available from https://portal.gdc.cancer.gov/ and https://cbioportal.org, and CPTAC images are available from https://wiki.cancerimagingarchive.net/display/Public/CPTAC+Pathology+Slide+Downloads. Annotations for HRD status are available in the published work of Knijnenburg et al. (47), and annotations for genomic ancestry were obtained from Carrot-Zhang et al. (46). Codes used for this analysis, trained models, and matched radiomic features/histology features from the UCMC validation dataset to replicate this analysis are available at https://doi.org/10.5281/zenodo.13785423; code is also available at https://github.com/fmhoward/HistoXGAN. Additional digital images can also be provided pending scientific review and a completed data use agreement. Requests for digital images should be submitted to F.M.H. (frederick.howard@uchospitals.edu). Licensing of code and data is through CC BY-NC 4.0.

Files

sciadv.adq0856.pdf

Files (8.7 MB)

Name Size Download all
Article
md5:d1571102df64344f615668cf7ef2d509
4.7 MB Preview Download
md5:e7cb322c646996efd2ef8ba5fb9b477b
3.9 MB Preview Download

Additional details

Identifiers

DOI
10.1126/sciadv.adq0856
Other
oai:uchicago.tind.io:14030

Funding

National Institutes of Health
R56DE030958
U.S. Department of Defense
BC211095P1
American Cancer Society
National Cancer Institute
K08CA283261
National Cancer Institute
R01CA276652
National Cancer Institute
P50-CA058223
Cancer Research Foundation
Lynn Sage Breast Cancer Foundation
ASCRS Research Foundation
BCRF-23-127
Stand Up To Cancer
Horizon Therapeutics
2021-SC1-BHC
Adenoid Cystic Carcinoma Research Foundation

UChicago Information

Division(s)
Biological Sciences Division
Department(s)
Medicine